Fix an N^2 in batch hydration

[bshepherdson]

This hadn't been noticed previously since N is generally small, eg. 50 or 60 at most. Working with a Metabase dashboard that loaded 10000 fields (100 each from 100 tables) in a single select + hydrate, I found this was consuming about 3 seconds.

The slow path was effectively a very slow conj: taking the acc vector, and then for each annotated-instances doing:

(recur (vec (concat acc nil [(first hydrated-instances)]))
       ...)

which is pouring the whole vector into a lazy sequence and back into a vector, once for each instance in the list.

The new version uses a custom transducer to do the same process in O(n) time (and without any seq overhead, as a bonus).

[camsaul]

Can you update the failing tests?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 83.58%. Comparing base (3729d20) to head (4748350).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #172   +/-   ##
=======================================
  Coverage   83.58%   83.58%           
=======================================
  Files          37       37           
  Lines        2498     2498           
  Branches      212      212           
=======================================
  Hits         2088     2088           
  Misses        198      198           
  Partials      212      212           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[alexander-yakushev]

Bump. I've rewritten the implementation of merging original instances with batch-hydrated instances again. I'm building an index that serves both as the indicator which instances should be taken into the batch to be hydrated and which not; and also it enumerates the instances-to-by-hydrated, so that on the second pass of this index, we can extract the hydrated instances by their position in the batch.

The tests are now fixed.